I think it depends on the learning goals and how the teacher frames the prompt to guide text and language analysis. If the concern is “errors”, learner corpora, at least in MACAWS, offer enough “error-free” examples that would maintain the writing produced by the learners untouched thus original so AI tools would not be necessary at least in this context and in the approach we subscribe to when teaching with learner corpora.
As a way to work around, the teacher could select only concordance lines where errors don’t appear, but we have encouraged the use of concordance lines with some errors so students can develop language awareness. The focus would not be on accuracy only but also on developing the skills of noticing and analysis so students can rewrite sentences using the standard rules, describe patterns, and justify choices in a way to expand from the usual modeling and mimicking a pattern. Of course this would be more useful after they have seen standard patterns so they have an initial reference to draw from, but again I think learner corpora do offer plenty examples of standard patterns which do not require “correction” by tools.